Entropy Minimization & Locating Faults Across the Electrical Network Using Customer No Light Calls

Abstract

This thesis develops a probability model that uses information from no light calls to estimate the probabilities of faults across the electrical distribution network. We first consider a single fault per circuit scenario, before developing the logic for the multiple fault scenario. Due to the problem of high dimensionality in the multiple fault logic, we cannot calculate the probabilities based on our assumptions exactly. Instead, we use Monte Carlo methods to estimate the edge fault probabilities. The Monte Carlo method for estimating probabilities is then implemented in the robust grid simulator, which allows us to run storm simulations and observe the results. Shannon entropy is introduced as a metric to measure our uncertainty about the edge fault probabilities over a circuit. In the last part of the thesis, we observe how changing major parameters, such as the no light call probability and setting of our prior, affects circuit entropies. We use two simple heuristic policies, pure exploitation and Boltzmann exploration to send lookup trucks to measure edges with the objective of minimizing circuit entropy.